How to capture footage that just works

You spun the phone around while standing still, instead of walking around the subject. The most common cause of 'few sparse points.'

3D reconstruction needs PARALLAX — the camera has to see the subject from different physical positions, like your two eyes seeing the world from slightly different points. Rotating in place gives the system the same view at different angles, which it can't triangulate into 3D.

• ·Walk slowly AROUND the subject — your feet should move, not just your wrists.
• ·For an object on a table: take a step every couple seconds, keeping the subject in the middle of the frame.
• ·For a room: walk a loop or zigzag through it, not just turn from a fixed spot.
• ·If you can only stand in one place, that's a panorama, not a splat. Splats need translation.

Blank walls, smooth floors, solid colors, ceilings, sky. The matcher needs distinctive features (corners, edges, patterns) to lock onto.

Every frame gets analyzed for ~hundreds of distinctive points the system can re-find in adjacent frames. Featureless surfaces give it nothing to chain together, so it ends up with too few matches to recover 3D positions.

• ·Pick subjects with visible texture — wood grain, fabric folds, bricks, foliage, surface patterns.
• ·Avoid filling the frame with one solid color (a plain wall, a sky-only shot, a white ceiling).
• ·If the subject itself is smooth, make sure the background or surroundings give the matcher something to lock onto.
• ·Glossy / glassy subjects are a separate problem — see below.

Blurry frames from quick pans, shaky hands, or low-light captures with slow shutter speeds.

Motion blur smears the fine details the system uses to identify each frame uniquely. A blurred edge looks like a different shape than the same edge sharp, so the matcher can't chain them together.

• ·Slow the camera pass. A 30-second smooth orbit beats a 10-second jerky one — every time.
• ·Hold the device with both hands and brace against your body, a wall, or anything stable.
• ·Avoid low light — your phone drops shutter speed and motion blur gets worse fast.
• ·If you have a gimbal, use it.

Reflective surfaces confuse the matcher. The reflection in a mirror or a glossy car panel moves when the camera moves, so it looks like a 3D point that doesn't actually exist.

Standard reconstruction assumes every visible feature is a real 3D point at a fixed location. Reflections break that assumption — the matcher tries to triangulate them and gets garbage 3D positions, which then poison the rest of the solution.

• ·For glassy subjects: enable GUT (ray-traced training) in your processing recipe. It handles reflections properly.
• ·Shoot under soft, diffuse light — direct sunlight or harsh overhead lights make every glossy edge a tiny mirror.
• ·For real mirrors: cover them, or pick a different angle that excludes them.
• ·Polarizing filters help on phones that support them.

You only covered part of the subject — one side, one angle, half a room. The system needs overlapping views from multiple angles to reconstruct complete geometry.

3D reconstruction is triangulation: every point in the final splat has to be visible from at least two (ideally three+) different camera positions. Anything you only saw from one angle either gets reconstructed badly or not at all.

• ·Make a complete loop around the subject.
• ·Keep the subject in frame the whole time — don't aim the camera at the ground or sky between angles.
• ·Aim for ~30% overlap between adjacent frames. If a feature is in frame N, it should still be in frame N+1.
• ·For 360° subjects: 60-120 frames over a full orbit is usually the sweet spot.